this post was submitted on 22 Jun 2023
3 points (100.0% liked)

tl;dr bot discussion

88 readers
1 users here now

Community for discussing tl; dr bot. Leave any feedback here.

founded 2 years ago
MODERATORS
3
How the bot works (self.tldrbot)
submitted 2 years ago* (last edited 2 years ago) by poopsmith to c/tldrbot
 

The bot just makes use of:

  • SMMRY - I'm evaluating this and might eventually switch over to some ml library
  • LemmyBot

As posts come in on a news community (via LemmyBot), I grab the url and pass it to smmry, and reply. And that's it.


Update June 22, 2023: I'll probably stop using SMMRY as it's pretty buggy and not usable ~20% of the time. I'm thinking about using a headless browser (like Selenium), grabbing the text, and using GPT to analyze and summarize the text. I need to try this out and guesstimate the pricing for it.

top 6 comments
sorted by: hot top controversial new old
[–] poopsmith 3 points 2 years ago

SMMRY kinda sucks when using the URL mode. It'll sometimes grab the gdpr cookie message and summarize that, or it'll grab weird metadata.

[–] botlt_test 2 points 2 years ago* (last edited 2 years ago)

I like the concept!, keep it up! Edit: us bots must unite!

[–] [email protected] 1 points 2 years ago

Two suggestions:

  1. you should change the "tl;dr" heading to say "tl;dr (machine-generated)" to avoid wasting people's time who know they don't want to read LLM-generated summaries.

  2. you should make the suggestions link be a relative link, like this: [Suggestions?](/c/[email protected]) so that it works for users on other instances too.

Tbh I would still lean toward banning this bot even if you take the above suggestions, because LLM-generated texts are often subtly wrong and basically worse than useless imo (eg, often actually harmful).

But making it more clearly labeled as being what it is would be an improvement.

I'm on the admin team at lemmy.ml and have banned a few people posting totally-not-labeled but obviously-LLM-generated comments lately, and I'm getting pretty tired of this automated bullshit tech. But since yours is at least somewhat labeled, and hopefully better labeled soon if you take my suggestions above, I won't ban this one without discussing with other admins first.

[–] Seven 1 points 2 years ago* (last edited 2 years ago)

The bot summarising long lemmy comments in 1 sentence per 150 characters after getting tagged maybe, where it summarizes the comment previous to the tag comment

[–] [email protected] 1 points 2 years ago

I’m thinking about using a headless browser (like Selenium), grabbing the text, and using GPT to analyze and summarize the text. I need to try this out and guesstimate the pricing for it.

There are plugins these days that you can make the LLM use to search, retrieve and extract content from the web. If you have OpenAI API access you can use that, or you could host an open model yourself (but I'm not sure what the situation is in regards to how well those handle plugin tool use).

[–] [email protected] 1 points 2 years ago

Hi, I see your bot is posting completely offtopic things like this - since it has a lot of replies I'll leave that one for posterity, but I'm banning it from [email protected] now. I'm not banning it from lemmy.ml yet, so (for now, at least) people here who appreciate it can still it its posts in [email protected] where I see you're also posting.